EP6: Learn How AI Innovation Is Becoming the Ultimate Competitive Advantage

00:00:08: Hi, I'm Michelle Booth CX AI lead for the Amir region at Zoom and i am thrilled to be joined by Zoom's head of product for CXAI Ram Rajagopalan.

00:00:19: Welcome Ram!

00:00:20: Hey Michelle nice to see you and thank you for this opportunity.

00:00:24: Great to have you here.

00:00:25: obviously AI is moving incredibly fast.

00:00:29: most organizations we speak with on a daily basis are not seen as challenging the amount of technology that's available.

00:00:37: It's more around how they're applying it in a way, really solves problems and delivers measurable value.

00:00:44: so today I'm really keen Ram to unpack

00:00:47: how

00:00:48: zoom on yourself as the head of product for CX AI building AI capabilities that are starting with real customer needs and turning those into practical scalable outcomes.

00:01:00: So let's start at When you're thinking about building a new AI capability, how do make sure it starts with real customer problems rather than technology for technology sake?

00:01:13: Yeah.

00:01:13: That's great question.

00:01:16: We are living in an era where things and AI moving at very rapid pace.

00:01:23: One of the thing that is fundamental to any product builder Is to stay close to customers' problem.

00:01:31: So one of the things that we do often is, uh... We don't start with what can AI do even when you're building our own products.

00:01:39: But we have meaningful partnerships and discussions with customers.

00:01:44: for example- You know?

00:01:46: We see customers coming to us saying traditional IVRs, press One for this, Press Two For That has been in place but it's not serving their needs anymore and people like to have natural conversations in order we get routed the right person, or even not reading the queue but getting answers served faster.

00:02:07: So these are some fundamental problems that we hear.

00:02:10: And then when you think about how can we serve this problem better by using technologies whether it's an LLM or other technology then that is the next thing that follows.

00:02:23: But always, one of things we follow not just me or other PMs at Zoom too We start with customer problem and what does... We try to go ask them why?

00:02:36: Why's it a problem?

00:02:37: Then you got root cause from there.

00:02:40: That keeps us grounded And keep less distracted around instead of focusing on what can LLMs do for you but rather than focus and focus on, What are the key problems?

00:02:52: What is a key metrics that we're trying to move.

00:02:53: And how does it overall bring in an outcome That can be measured both from our customers standpoint as well as from us From a product's standpoint?

00:03:02: Do know I've seen that In the time that i have been working For Zoom Is How close You Are As A Product Team To What customers are actually asking for and taking that customer insight forward.

00:03:15: How do you balance between?

00:03:17: Actually, customers are asking for things which might be not the shiny new exciting technology but like I say they're solving real world problems Between what's going on with the competition And whats coming out.

00:03:32: What is the latest trend?

00:03:33: how to get that balance right when your developing a roadmap?

00:03:39: Yeah thats another thought for a question.

00:03:43: See, one of the things I've been now at Zoom for eighteen months and One other thing that has gotten used to when it's kind of delights me in a pleasant way is what we call internally as zoom speed.

00:03:56: you know What typically takes Other organizations six months or A year to deliver?

00:04:05: What i have seen our incredible AI and engineering teams do is we speak in weeks and delivery time in months.

00:04:15: In fact, sometimes we do deliver so fast that customers are surprised and stunned.

00:04:24: they need to take some time to digest and absorb these features.

00:04:28: So We have a very fast delivery cycle when it comes to execution style And that comes from...we're still founder-led company with Eric and it's a startup kind of atmosphere in every team, at any moment because that is how Eric likes to operate.

00:04:47: But going back again you know the goal was find balance between... We do speak with number customers.

00:04:54: we need rationalize needs for different segments like large enterprises or mid-market and small business.

00:05:03: And we try to find a pattern that can benefit everybody, for example if you look at AI... If you look in the specific place large enterprises just don't want an AI solution but they do not want a solution with enough guardrails security transparency into how the AI systems operate.

00:05:30: Sometimes, you know when you're dealing with large language models it becomes a black box because You end up using the frontier model from maybe Anthropic or a chat GPT model.

00:05:41: So those are requirements that be here.

00:05:44: But when we think about those requirements It not only benefits one customer but we see that benefiting every segment of the customer Because Everybody wants AI not to go haywire.

00:05:55: They want security, guardrails, trustworthiness and scalability as the needs grow.

00:06:01: so we kind of pat it down and see what features make sense for everybody and kind of prioritize that.

00:06:07: And then you have this zoom velocity through which we deliver them within weeks.

00:06:13: This market is evolving fast.

00:06:14: We are always keeping our ear-to-the ground and understanding what exactly needs to be delivered for customers, to be successful within their organization as well.

00:06:22: That's one drives our not star motive.

00:06:26: It is really refreshing to hear that approach.

00:06:30: One of the capabilities have recently been launched with Zoom Virtual Agent Is a multi-modal AI OCR capability optical character recognition for anyone who is not familiar with that term.

00:06:47: But it's really interesting to get your view on what problems you see, the capability solving of customers and interacting with?

00:06:57: Yeah I think this was one of most exciting features we have launched recently.

00:07:04: speaking to customers and analysts everybody is thrilled about.

00:07:10: If you think about our daily lives, right?

00:07:12: We are all living in a multi-modal world.

00:07:16: Meaning that we're doing multitasking and talking to someone but at the same time texting somebody else.

00:07:22: what this feature brings too.

00:07:24: In the word of customer services You might be let's say for example Say got your new television delivered or have a router That is delivered But it isn't working well But then you're calling a customer support agent.

00:07:40: What this multi-modality does is we have an AI voice agent that can try to resolve these problems, and you can speak to the agent and describe your problem but even take our small photo of this and send it into the agent so that the virtual agent as powered by large language models can understand your context better.

00:08:04: And for example, let's say you need to send a serial number of your device instead of reading it out which could be cumbersome.

00:08:11: You can just take the photo and set up with the virtual agent so that they understand what product is or when was purchased whether within warranty.

00:08:21: The goal here again speed-up customer service right?

00:08:26: You don't want customers waiting but have them get the resolve, the problems that they're calling about.

00:08:37: So that's the goal and that's a use case if you are targeting with this kind of instance.

00:08:41: And we just getting started.

00:08:43: You know We starting reading text from images.

00:08:46: We can interpret images.

00:08:48: It is also bound by guardrails and safety protocols.

00:08:51: so...we keeping it in mind In future we might be able to support even short videos.

00:08:56: So send us small video and understand exactly.

00:09:00: The goal again is serve customers in shortest fast possible way without keeping them waiting and getting the answers they need.

00:09:07: That's so cool!

00:09:11: And I can completely see the value of that.

00:09:17: If you're trying to describe something, like say read out a long number it might be a serial number on your product or something on an invoice.

00:09:26: so yeah i could really see the values there.

00:09:28: but guess im also interested... You mentioned it could actually read what's in a photograph.

00:09:33: So lets say for example I'm letting you know about something that's damaged, maybe i've ordered something it's arrived damage and try to send your photo but...I accidentally sent you a photo of my dogs.

00:09:46: Would the virtual agent know?

00:09:50: Yeah again we have taken enough guardrails to know.

00:09:55: ultimately we need to classify their image into one of them is relevant for the conversation, you might be talking to an insurance agent but he sent a wrong photo.

00:10:05: So it has intelligence enough to know.

00:10:08: is this contextual to the conversation?

00:10:11: If not ,it will prompt you say hey!

00:10:13: This image that we send doesn't seem to have context for this conversation.

00:10:17: maybe isn't the wrong image and let prompt you again.

00:10:20: so yeah That's part of the agentic platform understands these contexts better And lets give opportunity back another image.

00:10:31: So yeah, this along with guardrails too.

00:10:33: Like you know in terms of these are right images.

00:10:36: it's what are the guardrail specific to each brand?

00:10:40: And much more than that there is one thing which I think most people kind forget is reporting and analytics.

00:10:47: so the images that we send now becomes part of engagement metadata.

00:10:54: if somebody from the organization want do analysis on post call.

00:10:59: All of this is stored and accessible to the person within the organization, to do post-call analysis.

00:11:05: To understand how we can improve customer experience better?

00:11:10: So it's all part of the Zoom platform that retains that across the board.

00:11:14: Yeah I see the value certainly from a customer perspective but absolutely form an organisational perspective.

00:11:21: being able just get information check through automation and AI which are accurate upload that into the right place, it's going to save a lot of rework.

00:11:31: But I think the crucial thing here is making it more accessible.

00:11:36: bringing in this multi-modality makes the conversation more accessible.

00:11:41: and another big challenge that you know A lot of our customers And i think outside of the customers we work with anyone who was kind of exploring voice agents.

00:11:51: chat agents In terms of AI capabilities recognise that customers want to speak in their own language.

00:12:01: It's easy to explain something, you know?

00:12:03: In a language that you're familiar with rather than forcing people speaking at certain languages we can support.

00:12:10: We do a lot of languages and we've got lots of languages currently covered.

00:12:15: When thinking about prioritisation for language and trying make things more accessible How again do you prioritize those different languages and capabilities so that it delivers the right outcomes for customers we're serving.

00:12:31: Yeah, again a great question.

00:12:34: see one of things about.

00:12:35: Zoom is a global company brand known across the board or any corner in the world.

00:12:42: people recognise their brands.

00:12:43: with that you get responsibility to serve customers where they are.

00:12:53: So today, Zoom Virtual Agent both on chat and voice supports more than thirty languages.

00:12:59: In the US for example we have forty percent of conversations that happen is in Spanish.

00:13:06: So we do have support for Spanish language within

00:13:10: U.S.,

00:13:11: but at the same time, both on chat and voice languages like Japanese and European languages like French German Italian etc... so that is part of our brand value to support these languages across-the-board and serve customers in different regions.

00:13:32: But when we do serve these languages, even a few years ago the rolling out and adding these languages takes an enormous amount of effort into getting it right.

00:13:47: We need to make sure that language is not only accurate but also speaks to cultural values.

00:13:56: We do take enormous care in terms of how we roll out, how we test these languages both using human agents as well as automated testing.

00:14:07: So we can test and verify that the languages that we rollout have highest level quality but at the same time we prioritize where we see most demand especially for example In EU.

00:14:22: a lot of demand is for French or German We see a lot of demand for Japanese and then now we are going to sing.

00:14:29: A lot of the man for Indian regional languages, including Hindi Tamil

00:14:33: etc.,

00:14:34: so we have a motion to roll out these languages gradually.

00:14:38: but there's other part of this is many companies, including zoom.

00:14:42: For example Zoom uses zoom virtual agent for our own customer support The source of knowledge through which the which powers the virtual agent To answer questions it often in one language.

00:14:54: It's not always localized into different languages, so it is challenging for an organization to translate and maintain local copies of these various languages.

00:15:06: So we take that button off from the customer And the virtual agent actually translates.

00:15:13: if you have a support article on FAQ document which might be English We can use that then answer this in Spanish without asking.

00:15:23: So it reduces the burden of management from our customer standpoint, but at the same time we can support multiple languages quite easily.

00:15:32: In an upcoming release maybe in by end-of-quarter two We will have the ability to automatically detect languages dynamically.

00:15:40: You might start off in English But let's say midway you decide to switch to Spanish so that The virtual agent will also change dynamically through that language detecting your language of preference during the call and dynamically switch to that language, speak back to you in that language.

00:15:57: And I think again... The goal is to provide better customer experience and better customer satisfaction.

00:16:04: That's what drives these kind-of languages to allow within Zoom

00:16:09: So powerful!

00:16:11: It truly meets customers where they are You know?

00:16:16: People who can speak multiple languages.

00:16:22: you know, being able to think and in the moment having this change.

00:16:27: the language of speaking is probably something that happens quite a lot.

00:16:32: And it's certainly something I've heard customers asking for, you know what can it automatically detect?

00:16:37: and so it's brilliant to know that's coming very excited about that capability.

00:16:42: That would be my superpower if i could have some power.

00:16:45: It will be able to understand any language.

00:16:47: So I'm going get through using virtual agent.

00:16:52: Now one thing I think we both recognise is super important with AI, it's the role of humans.

00:17:00: So we're doing a lot with AI to not just support the customer but support human agents like translation prompting them with relevant knowledge even in real time escalating when a sentiment changes so that we can alert a supervisor or team.

00:17:20: how does their human loop approach help?

00:17:23: improve virtual agents over time.

00:17:27: Yeah, yeah that's again another good question we often hear from our customers too.

00:17:34: is this feedback loop?

00:17:36: Is there a feedback loop that helps improve the Virtual Agent?

00:17:39: See in reality any Virtual Agent you know limits to what it can essentially solve for.

00:17:52: There are limits that the brands themselves have.

00:17:55: They want the virtual agent to solve certain problems, and they want human agents to solve different problems.

00:18:00: So what we do is when a customer calls in And the Virtual Agent tries To Solve The Problem It realizes That it may not Have The Necessary Skills Or Its Empowered To Solves Their Problems.

00:18:15: So It Decides To Escalate or Transfer The Call To A Human Agent.

00:18:20: And then the human agent, you know sometimes they have tribal knowledge.

00:18:24: They have knowledge that is not in any training manuals.

00:18:27: That's what Human Agents are like.

00:18:30: accumulated a lot of knowledge over the period many years but it isn't essentially documented It's all in their head.

00:18:40: So because we're part of Zoom ecosystem and platform We have disability into conversations.

00:18:47: that happens within virtual agent, but also when the call goes to a human agent.

00:18:51: Whether they're using expert assist or not we do see that conversation.

00:18:56: and then this is where Human in The Loop comes into play.

00:18:59: And what we recently introduced was this Expert feedback.

00:19:04: Where?

00:19:05: We look at calls That were sent To A Human Agent.

00:19:09: When you Look At the Analyze Those Transcripts & See Are There Things Nuggets Of Information that can be fed back to the virtual agent.

00:19:17: so in future somebody calls with a similar problem, we empower them solve those problems for customers.

00:19:26: We do a post-call analysis of these calls elevated to human agents and see if there is a corresponding knowledge article exists.

00:19:38: we kind of nudge the supervisor or the administrator to say, here are the topics that virtual agent could not answer.

00:19:45: And they have created this knowledge article through which it can answer these calls and conversations in the future.

00:19:51: would you approve?

00:19:52: This is up to the supervisor, the administrator at a company.

00:19:58: No, I don't want you to continue answering because some brands have sensitivity on certain topics that they do not want the virtual agent.

00:20:07: They are in full control and we had feedback loop coming back.

00:20:11: so this is a unique feature which differentiates Zoom Virtual Agents against many of our competition.

00:20:20: That's great capability making the virtual agent smarter.

00:20:25: But I love the fact that they're controlled with humans, because like you say not everything is appropriate to be a virtual conversation.

00:20:32: so we can surface this.

00:20:35: how are you solving and if it's been solved in the scenario where there hasn't been written down or recorded process then every human agent could use a new procedural a piece of content based on what we've surfaced.

00:20:54: Yeah,

00:20:56: so just to get in terms of closing because it's been really insightful quick update Ram.

00:21:04: but if we were to have this conversation again say two years from now What do you think?

00:21:11: Will have the biggest impact and how organizations will be serving their customers?

00:21:18: That's uh I wish i had the crystal ball to answer.

00:21:22: And so that's kind of a thing.

00:21:23: I mean, things are changing so dramatically.

00:21:28: even in the month before we didn't have this anthropic came up with open claw... Things were changing very dramatically.

00:21:42: So i don't know what two years from now but do see this right?

00:21:45: Do you see What We Are Trying To Do?

00:21:49: AI, our contact center AI which I'm privileged to be part of and an opportunity lead.

00:21:56: We are in the business of creating productivity surplus right?

00:22:02: What i mean by that is when we use virtual agents or when you use expert assist agents Right!

00:22:16: in a more delightful way, previously which they couldn't.

00:22:22: Even Zoom support for example.

00:22:24: we get ten thousand calls a month approximately and you know previously had only limited capacity with our human agents to answer these calls but as we introduce virtual agents it frees up the time for their Human Agents do more complex tasks that they didn't have the time to do But also gives us an opportunity to answer every single call That comes from a zoom customer.

00:22:46: So thats where I think about It.

00:22:48: you're going to see in the next few, every company in space as they deploy AI.

00:22:55: You are gonna create this productivity surplus.

00:22:58: now.

00:22:58: how does those productivity surplus gets used?

00:23:01: is up to the brand and up to any customer right?

00:23:04: I do feel fundamentally that customer experience today it's not flawless there so many things don't have much room for improvement.

00:23:15: So what i think might happen?

00:23:18: How do as a society, how do we use this productivity surplus?

00:23:22: Do we take the surplus and serve more people.

00:23:25: That is question for us to answer collectively but my job Is to provide My customers, serve my customers with more productivity And give them that option To decide how they used their productivity within the organisation.

00:23:40: Yeah I think it's something you said there Definitely what i'm hearing.

00:23:46: the complexity of a live agent role is increasing, because the AI is dealing with a lot of their basic queries now.

00:23:55: And so I think there's definitely going to be room for how we continue to help live agents become super-agents being able to do more...

00:24:07: More complex things?

00:24:08: Yeah,

00:24:08: fewer customers but more complex and more emotive.

00:24:12: Ram, just want say Thank you so much for sharing your insight on how the Zoom product team are working towards realising customer ambitions with their capabilities that they're delivering.

00:24:26: I really enjoyed this conversation and thank-you very much.

00:24:30: Same Michelle, thanks for the opportunity.

00:24:32: always a pleasure talking to you.

00:24:33: now let's get back to our customer call!

00:24:36: Absolutely.